A Nonintrusive Parametrized Reduced-Order Model for Periodic Flows Based on Extended Proper Orthogonal Decomposition

The periodic flows, such as vortex shedding and rotating flow in turbomachinery, are very common in both scientific and engineering fields. However, high-fidelity numerical simulations of unsteady flows are usually time-consuming, particularly when varying flow parameters need to be considered. In this paper, a novel nonintrusive parametrized reduced order model (PROM) approach for prediction of periodic flows is presented. The establishment of this ROM is based on two techniques, proper orthogonal decomposition (POD) and discrete Fourier transform (DFT), where the first one can extract the spatial features and the second has the ability to quantify the temporal effects of parameters. A prediction model based on artificial neural networks (ANNs) is used to map the flow parameters with DFT coefficients. Flows past a cylinder and two dimensions turbine flows are used to demonstrate the effectiveness of the proposed PROM. It is shown that the proposed POD-DFT-ANN (PDA) ROM are both efficient and accurate for the predictions of periodic flows with varying flow parameters.